A LoRa representing the concept of "Self Breast Sucking" and "Nipple licking".
Request form is closed temporarily as I'm busy IRL and i can't process all the incoming requests, I'm gonna still do the request submited before the form closed.
You can check the status of the loras i've been requested here
If you want to support me buy me a coffee (commisions opening soon)
I wanted to make a concept of this because the ones I tried are either overfit and generate bad results or have style bleed (In some case both), this would make it my first concept done so now I've got more knowledge on how they work, you can expect more concepts in the future :)
Details
Trigger: jelobreastsuck
Default:
trigger, self breast sucking
Licking nipple:
trigger, self breast sucking, tongue (You can add "licking nipple" aswell but not needed in most cases)
Double breast suck (Hard to get):
trigger, self breast sucking (In some cases adding "double breast sucking" can help)
Default but without hands support (Hard to get):
trigger, self breast sucking, <other poses for the arms/hands> ("arms behind head" tends to make it work more frequently for what I've tested)
Works on: 0.7-1
Tested angles and views:
from side: Works well in most cases
from below/from above: Tends to work a little better than from side, so it works good too
on back/lying: Works good aswell
on side/on stomach: Requires width > height (i.e: 768x512) to work well. Needs some more effort but not much
Tips:
1.Use a checkpoint that can do nsfw quite easily to make things easier (works good on non nsfw models aswell for what I've tested, maybe not on all of them)
2.Try to not include prompts that can obstruct the pose, if the character is clothed try to use "nipples" and/or "open clothes" on the prompt
3.If your using character loras and it doesn't work try to lower the weight of the the character's lora
4.Using sizes too big for the breasts it can add a third breast but it works (I tried with huge breasts:1.6)
You won't get perfect results each try, it depends on many things like the pose of the character, prompt, type of clothing, breast size, etc but for what I've tested maximum takes 5 tries to get a good result.
If you need help post a comment and I'll gladly try to help :D
If you liked it please rate the model and do a post, I enjoy seeing other's creations, and follow me if you want to get notifications about upcoming models.
Description
FAQ
Comments (16)
try claim model in tensor art bro
hi, i want to ask about training lora, how mutch value do put in network rank and alpha??
It depends on what i train, what are you trying to do?
@Jelosus1 im trying to train character , but i want the lora to take the model style,
@fuinypain you mean you want the character lora to have the same style as the checkpoint?
Usually 16 network and 8 dim is used, you can lower it down too but doesn't matter. If the LoRA have more than single character in it it still works with 16/8 but doesn't hurt to go for 32/16.
As for the your reply where you want lora to take the model style, I'm 100% sure you're overbaking the LoRA. I'd recommend you to go through holostrawberry's guide on LoRA making here. It makes use of collab, but the instructions apply to all trainers that are using kohya.
Yea, what wrench said it's correct, holo's guide is very good, if you need any help you can contact me on Discord :)
thank you guys ♥@Jelosus1 @wrench1815
what are u doing my guy
What others failed to do :goldTnp:
what vae do you use in the sample images?
clearvae_main, it's baked in my mix
@Jelosus1 thank you